Spectrum Sharing With Decentralized Occupation Control in Rule Regulated Networks

Abstract

Decentralized dynamic spectrum allocation (DSA) that exploits adaptive antenna array interference mitigation diversity at the receiver, is studied for interference-limited environments with high level of frequency reuse. The system consists of base stations (BSs) that can optimize uplink frequency allocation to their user equipments (UEs) to minimize impact of interference on the useful signal, assuming no control over resource allocation of other BSs sharing the same bands. To this end, "good neighbor" (GN) rules allow effective trade-off between the equilibrium and transient decentralized DSA behavior if the performance targets are adequate to the interference scenario. In this paper, we 1) extend the GN rules by including a spectrum occupation control that allows adaptive selection of the performance targets; 2) derive estimates of absorbing state statistics that allow formulation of applicability areas for different DSA algorithms; 3) define a semianalytic absorbing Markov chain model and study convergence probabilities and rates of DSA with occupation control including networks with possible partial breaks of the GN rules. For higher-dimension networks, we develop simplified search GN algorithms with occupation and power control and demonstrate their efficiency by means of simulations.

Good Neighbor Alternative to Best Response and Machine Learning Based Beamforming and Power Adaptation for MIMO Ad Hoc Networks

“…To investigate the convergence and equilibrium properties of the algorithms in Section III, we expand the semi analytic absorbing Markov chain model from [10], [11] to the considered beamforming and power allocation scenario. Semi-analytic (analytic for the given network configuration and propagation channel realizations) Markov chain modeling allows investigation of networks, when convergence to NE (absorbing states in Markov chain terminology) cannot be guaranteed with probability one.…”

“…This illustrates performance trends and applicability areas for different algorithms that can be further verified for lager scale networks by means of simulations. Markov chain analysis is used in [10], [11] to study a trade off between equilibrium and transient performance for the GN and selfish algorithms. In the considered problem with the fixed performance targets, we use Markov chain modeling in a different way to study probabilities of absorption and transitions in different groups of states, which allows us to predict desirable/undesirable convergence behaviour leading to the equilibrium or "voluntary dropout".…”

“…, I, which completely defines the Markov model of the considered spectrum sharing network. Then, for the given scenario realization, algorithm and its parameters, the number of absorbing states (AS) and ergodic subchains (ES), the equilibrium performance, the non-convergence probabilities, and the average number of sensing intervals before absorption can be found similarly to [11] using the states classification algorithm from [16].…”

“…This allows a trade-off between the equilibrium and transient performance leading to controllable convergence behavior without any explicit communications between coexisting subsystems. Application of GN self-organization is addressed in decentralized spectrum sharing in cognitive radio networks [10], [11], and distributed beam scheduling in coexisting multi-RAT networks [12].…”